Optimal bandwidth allocation for hybrid Video-on-Demand streaming with a distributed max flow algorithm

Abstract

Video-on-Demand (VoD) streaming counts for the largest share of Internet traffic, and commonly relies on P2P–CDN hybrid systems. In such hybrid systems, a peer’s upload bandwidth utilization is critical to P2P mode, thus the bandwidth allocation algorithm is important to the performance of VoD streaming. The current research either makes impractical assumptions, or is inefficient in real scenarios. Meanwhile some industrial practice such as additive-increase/multiplicative-decrease (AIMD) heuristic, although effective in practice, lacks theoretical foundation. This paper develops an optimal bandwidth allocation algorithm for hybrid VoD streaming. Specifically, we propose a novel Demand Driven Max-Flow Formulation, which treats each peer’s bandwidth demand as the flow commodity. The proposed distributed Free-for-All Push–Lift algorithm can solve the formulation in each peer, and is free of any lock, shared memory and atomic operation. Following the theoretical analysis, we implement the algorithms in real-world VoD streaming systems. Through extensive evaluations we demonstrate that our approach can provide high-quality bandwidth allocation for hybrid VoD systems.